annotate snow.h @ 12475:9fef0a8ddd63 libavcodec

Move mm_support() from libavcodec to libavutil, make it a public function and rename it to av_get_cpu_flags().
author stefano
date Wed, 08 Sep 2010 15:07:14 +0000
parents 0f0cd6b5791f
children
Ignore whitespace changes - Everywhere: Within whitespace: At end of lines:
rev   line source
3198
6b9f0c4fbdbe First part of a series of speed-enchancing patches.
gpoirier
parents:
diff changeset
1 /*
6b9f0c4fbdbe First part of a series of speed-enchancing patches.
gpoirier
parents:
diff changeset
2 * Copyright (C) 2004 Michael Niedermayer <michaelni@gmx.at>
6b9f0c4fbdbe First part of a series of speed-enchancing patches.
gpoirier
parents:
diff changeset
3 * Copyright (C) 2006 Robert Edele <yartrebo@earthlink.net>
6b9f0c4fbdbe First part of a series of speed-enchancing patches.
gpoirier
parents:
diff changeset
4 *
3947
c8c591fe26f8 Change license headers to say 'FFmpeg' instead of 'this program/this library'
diego
parents: 3582
diff changeset
5 * This file is part of FFmpeg.
c8c591fe26f8 Change license headers to say 'FFmpeg' instead of 'this program/this library'
diego
parents: 3582
diff changeset
6 *
c8c591fe26f8 Change license headers to say 'FFmpeg' instead of 'this program/this library'
diego
parents: 3582
diff changeset
7 * FFmpeg is free software; you can redistribute it and/or
3198
6b9f0c4fbdbe First part of a series of speed-enchancing patches.
gpoirier
parents:
diff changeset
8 * modify it under the terms of the GNU Lesser General Public
6b9f0c4fbdbe First part of a series of speed-enchancing patches.
gpoirier
parents:
diff changeset
9 * License as published by the Free Software Foundation; either
3947
c8c591fe26f8 Change license headers to say 'FFmpeg' instead of 'this program/this library'
diego
parents: 3582
diff changeset
10 * version 2.1 of the License, or (at your option) any later version.
3198
6b9f0c4fbdbe First part of a series of speed-enchancing patches.
gpoirier
parents:
diff changeset
11 *
3947
c8c591fe26f8 Change license headers to say 'FFmpeg' instead of 'this program/this library'
diego
parents: 3582
diff changeset
12 * FFmpeg is distributed in the hope that it will be useful,
3198
6b9f0c4fbdbe First part of a series of speed-enchancing patches.
gpoirier
parents:
diff changeset
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
6b9f0c4fbdbe First part of a series of speed-enchancing patches.
gpoirier
parents:
diff changeset
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
6b9f0c4fbdbe First part of a series of speed-enchancing patches.
gpoirier
parents:
diff changeset
15 * Lesser General Public License for more details.
6b9f0c4fbdbe First part of a series of speed-enchancing patches.
gpoirier
parents:
diff changeset
16 *
6b9f0c4fbdbe First part of a series of speed-enchancing patches.
gpoirier
parents:
diff changeset
17 * You should have received a copy of the GNU Lesser General Public
3947
c8c591fe26f8 Change license headers to say 'FFmpeg' instead of 'this program/this library'
diego
parents: 3582
diff changeset
18 * License along with FFmpeg; if not, write to the Free Software
3198
6b9f0c4fbdbe First part of a series of speed-enchancing patches.
gpoirier
parents:
diff changeset
19 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
6b9f0c4fbdbe First part of a series of speed-enchancing patches.
gpoirier
parents:
diff changeset
20 */
6b9f0c4fbdbe First part of a series of speed-enchancing patches.
gpoirier
parents:
diff changeset
21
7760
c4a4495715dd Globally rename the header inclusion guard names.
stefano
parents: 5830
diff changeset
22 #ifndef AVCODEC_SNOW_H
c4a4495715dd Globally rename the header inclusion guard names.
stefano
parents: 5830
diff changeset
23 #define AVCODEC_SNOW_H
3198
6b9f0c4fbdbe First part of a series of speed-enchancing patches.
gpoirier
parents:
diff changeset
24
6b9f0c4fbdbe First part of a series of speed-enchancing patches.
gpoirier
parents:
diff changeset
25 #include "dsputil.h"
11485
0f0cd6b5791f Separate DWT from snow and dsputil
mru
parents: 11460
diff changeset
26 #include "dwt.h"
3198
6b9f0c4fbdbe First part of a series of speed-enchancing patches.
gpoirier
parents:
diff changeset
27
6b9f0c4fbdbe First part of a series of speed-enchancing patches.
gpoirier
parents:
diff changeset
28 #define MID_STATE 128
6b9f0c4fbdbe First part of a series of speed-enchancing patches.
gpoirier
parents:
diff changeset
29
6b9f0c4fbdbe First part of a series of speed-enchancing patches.
gpoirier
parents:
diff changeset
30 #define MAX_PLANES 4
6b9f0c4fbdbe First part of a series of speed-enchancing patches.
gpoirier
parents:
diff changeset
31 #define QSHIFT 5
6b9f0c4fbdbe First part of a series of speed-enchancing patches.
gpoirier
parents:
diff changeset
32 #define QROOT (1<<QSHIFT)
6b9f0c4fbdbe First part of a series of speed-enchancing patches.
gpoirier
parents:
diff changeset
33 #define LOSSLESS_QLOG -128
5587
3ae03eacbe9f use 16bit IDWT (a SIMD implementation of it should be >2x faster then with
michael
parents: 5565
diff changeset
34 #define FRAC_BITS 4
3314
aea2230e6033 Snow multiple reference frames
lorenm
parents: 3223
diff changeset
35 #define MAX_REF_FRAMES 8
3198
6b9f0c4fbdbe First part of a series of speed-enchancing patches.
gpoirier
parents:
diff changeset
36
3206
c1add9fe5c65 Snow mmx + sse2 part 2
corey
parents: 3198
diff changeset
37 #define LOG2_OBMC_MAX 8
3198
6b9f0c4fbdbe First part of a series of speed-enchancing patches.
gpoirier
parents:
diff changeset
38 #define OBMC_MAX (1<<(LOG2_OBMC_MAX))
6b9f0c4fbdbe First part of a series of speed-enchancing patches.
gpoirier
parents:
diff changeset
39
3223
8f048c3295ff altivec support for snow
lu_zero
parents: 3206
diff changeset
40 /* C bits used by mmx/sse2/altivec */
8f048c3295ff altivec support for snow
lu_zero
parents: 3206
diff changeset
41
5587
3ae03eacbe9f use 16bit IDWT (a SIMD implementation of it should be >2x faster then with
michael
parents: 5565
diff changeset
42 static av_always_inline void snow_interleave_line_header(int * i, int width, IDWTELEM * low, IDWTELEM * high){
3223
8f048c3295ff altivec support for snow
lu_zero
parents: 3206
diff changeset
43 (*i) = (width) - 2;
8f048c3295ff altivec support for snow
lu_zero
parents: 3206
diff changeset
44
8f048c3295ff altivec support for snow
lu_zero
parents: 3206
diff changeset
45 if (width & 1){
8f048c3295ff altivec support for snow
lu_zero
parents: 3206
diff changeset
46 low[(*i)+1] = low[((*i)+1)>>1];
8f048c3295ff altivec support for snow
lu_zero
parents: 3206
diff changeset
47 (*i)--;
8f048c3295ff altivec support for snow
lu_zero
parents: 3206
diff changeset
48 }
8f048c3295ff altivec support for snow
lu_zero
parents: 3206
diff changeset
49 }
8f048c3295ff altivec support for snow
lu_zero
parents: 3206
diff changeset
50
5587
3ae03eacbe9f use 16bit IDWT (a SIMD implementation of it should be >2x faster then with
michael
parents: 5565
diff changeset
51 static av_always_inline void snow_interleave_line_footer(int * i, IDWTELEM * low, IDWTELEM * high){
3223
8f048c3295ff altivec support for snow
lu_zero
parents: 3206
diff changeset
52 for (; (*i)>=0; (*i)-=2){
8f048c3295ff altivec support for snow
lu_zero
parents: 3206
diff changeset
53 low[(*i)+1] = high[(*i)>>1];
8f048c3295ff altivec support for snow
lu_zero
parents: 3206
diff changeset
54 low[*i] = low[(*i)>>1];
8f048c3295ff altivec support for snow
lu_zero
parents: 3206
diff changeset
55 }
8f048c3295ff altivec support for snow
lu_zero
parents: 3206
diff changeset
56 }
8f048c3295ff altivec support for snow
lu_zero
parents: 3206
diff changeset
57
5587
3ae03eacbe9f use 16bit IDWT (a SIMD implementation of it should be >2x faster then with
michael
parents: 5565
diff changeset
58 static av_always_inline void snow_horizontal_compose_lift_lead_out(int i, IDWTELEM * dst, IDWTELEM * src, IDWTELEM * ref, int width, int w, int lift_high, int mul, int add, int shift){
3223
8f048c3295ff altivec support for snow
lu_zero
parents: 3206
diff changeset
59 for(; i<w; i++){
8f048c3295ff altivec support for snow
lu_zero
parents: 3206
diff changeset
60 dst[i] = src[i] - ((mul * (ref[i] + ref[i + 1]) + add) >> shift);
8f048c3295ff altivec support for snow
lu_zero
parents: 3206
diff changeset
61 }
8f048c3295ff altivec support for snow
lu_zero
parents: 3206
diff changeset
62
8f048c3295ff altivec support for snow
lu_zero
parents: 3206
diff changeset
63 if((width^lift_high)&1){
8f048c3295ff altivec support for snow
lu_zero
parents: 3206
diff changeset
64 dst[w] = src[w] - ((mul * 2 * ref[w] + add) >> shift);
8f048c3295ff altivec support for snow
lu_zero
parents: 3206
diff changeset
65 }
8f048c3295ff altivec support for snow
lu_zero
parents: 3206
diff changeset
66 }
8f048c3295ff altivec support for snow
lu_zero
parents: 3206
diff changeset
67
5587
3ae03eacbe9f use 16bit IDWT (a SIMD implementation of it should be >2x faster then with
michael
parents: 5565
diff changeset
68 static av_always_inline void snow_horizontal_compose_liftS_lead_out(int i, IDWTELEM * dst, IDWTELEM * src, IDWTELEM * ref, int width, int w){
3223
8f048c3295ff altivec support for snow
lu_zero
parents: 3206
diff changeset
69 for(; i<w; i++){
5565
93082c591c8b Change rounding of the horizontal DWT to match the vertical one.
michael
parents: 5552
diff changeset
70 dst[i] = src[i] + ((ref[i] + ref[(i+1)]+W_BO + 4 * src[i]) >> W_BS);
3223
8f048c3295ff altivec support for snow
lu_zero
parents: 3206
diff changeset
71 }
8f048c3295ff altivec support for snow
lu_zero
parents: 3206
diff changeset
72
8f048c3295ff altivec support for snow
lu_zero
parents: 3206
diff changeset
73 if(width&1){
5565
93082c591c8b Change rounding of the horizontal DWT to match the vertical one.
michael
parents: 5552
diff changeset
74 dst[w] = src[w] + ((2 * ref[w] + W_BO + 4 * src[w]) >> W_BS);
3223
8f048c3295ff altivec support for snow
lu_zero
parents: 3206
diff changeset
75 }
8f048c3295ff altivec support for snow
lu_zero
parents: 3206
diff changeset
76 }
8f048c3295ff altivec support for snow
lu_zero
parents: 3206
diff changeset
77
7760
c4a4495715dd Globally rename the header inclusion guard names.
stefano
parents: 5830
diff changeset
78 #endif /* AVCODEC_SNOW_H */